Pixie-Anim

A high-performance, zero-dependency short-form animation optimizer written in Rust and WebAssembly.

Inspired by Lee Robinson's pixo, Pixie-Anim follows a similar philosophy: building mission-critical codecs from scratch without heavy runtime dependencies, optimized for both native CLI performance and zero-latency in-browser execution.

• 📖 Read more: Optimizing GIFs with Pixo: Extending Lessons Learned - A technical writeup on the architecture and philosophy behind Pixie-Anim.

🚀 Key Features

• Zero-Dependency Core: From-scratch implementation of GIF89a and LZW.
• Superior Compression: Achieves 30-50% smaller files than Gifsicle -O3 through advanced heuristics.
• Perceptual Quantization: Uses CIELAB color space and K-Means++ for superior color fidelity.
• Temporal Denoising: Index-reuse logic that treats temporal noise as negligible, dramatically reducing size for real-world video.
• Stable Dithering: Choice of Ordered (Bayer) or Blue Noise dithering to eliminate spatial "shimmer."
• Evaluation-Driven Presets: Smart High/Medium/Low presets determined through thousands of iterations of Gemini-based perceptual scoring.
• Parallelized & SIMD: Multi-threaded quantization via Rayon and Planar AVX2/NEON kernels for nearest-color search.
• WASM Playground: Fully functional, client-side optimizer with direct MP4-to-GIF support.
• Subjective LLM Judge: Automated quality evaluation via Gemini 3 Flash.

📊 Performance Matrix (Space Waves - 720p)

Benchmarked on the high-complexity "Space Waves" sequence (Veo 3.1). Pixie-Anim provides a massive size advantage while improving visual stability through Ordered Dithering.

🧠 Algorithms

Zeng Palette Reordering

By ordering the 256-color palette by visual similarity, we maximize the length of LZW strings, gaining "free" compression with negligible compute cost.

Lossy LZW (Fuzzy Neighbor Matching)

A unique optimization where the encoder can match visually similar "neighbor" indices in the Zeng-ordered palette to continue an LZW sequence, further collapsing file size.

Planar SIMD Search

We discovered that standard interleaved RGB SIMD is often slower than scalar for small palettes. Pixie-Anim uses a Planar data layout to unlock the true potential of AVX2/NEON for nearest-color search.

🛠 Usage

CLI

Build the native binary:

cargo build --release --features="cli"

The binary will be available at target/release/pixie-anim.

Optimize an image sequence:

./target/release/pixie-anim frames/*.png --output optimized.gif --fps 15 --lossy 8 --fuzz 10

Web Playground

The web UI is a Lit-based application that leverages the Rust core via WASM.

1. cd web
2. pnpm install
3. pnpm run dev (This automatically triggers build-wasm to generate bindings)
4. Drop an MP4 or GIF into the browser and select an Optimization Preset (Low, Medium, High).
5. Pixie-Anim will process the file locally in your browser.

Note: To prevent browser memory exhaustion, the web playground currently scales video to 640px and caps extraction at 300 frames. Use the CLI for unlimited high-resolution processing.

📊 Benchmarking & Evaluation

Pixie-Anim includes a unified benchmarking tool to compare performance and visual quality against Gifsicle, FFmpeg, and gifski.

See tests/README.md for detailed instructions on the Hill-Climbing workflow and iterative performance tuning.

Quick Benchmark:

./target/release/pixie-bench --input video.mp4 --name my_test --lossy 8 --report tests/benchmarks.md

This tool automatically:

1. Extracts frames via ffmpeg.
2. Runs the optimization suite across all four engines.
3. Invokes the Gemini Subjective Judge for quality scoring.
4. Generates a detailed Markdown report with size, speed, and visual critique.

👁️ Automated Evaluation

Pixie-Anim includes a judge tool that uses Gemini 3 Flash to perform a comparative analysis of the output, identifying artifacts like banding or temporal jitter that traditional metrics (PSNR/SSIM) might miss.

📋 Prerequisites

Core Development

• Rust (1.70+)
• Node.js & npm/pnpm (For Web Playground)
• wasm-bindgen-cli (cargo install wasm-bindgen-cli)

Benchmarking & Evaluation

To run the comparative benchmarks and the subjective judge, you'll need:

• FFmpeg: For frame extraction and judge frame analysis.
• Gifsicle: For baseline compression comparison.
• gifski: For high-quality comparison.
• bc: For benchmark timing calculations.
• Gemini API Key: Set as GEMINI_API_KEY in your environment or .env file for the judge tool.

Install on macOS via Homebrew:

brew install ffmpeg gifsicle gifski bc

Part of the Pixie family of high-performance media tools